0.1 Abstract

questions for abstract style: voice (passive vs active, ‘we’ vs ‘this paper’), tense (present vs past)…

Two global datasets of species distribution maps - AquaMaps and the International Union for Conservation of Nature - dominate our understanding of marine species ranges throughout the world’s oceans, and inform a wide range of biodiversity and conservation purposes. Differences in methodology and intent between these two datasets drive differences in predicted species ranges, with significant implications for management and conservation decisions.

Together, the two datasets provide range information for XXX species, with only 2166 species common to both datasets. what is the key takeaway for the taxonomic and spatial coverage analyses? can we say something about benefits of combining the two for better taxo coverage? For the subset of species included in both datasets, we examine the alignment of range maps by comparing range distribution and extent. Categorizing the results based on quality of alignment, we identify several problems? that commonly lead to disagreement in species range predictions. Notably, IUCN maps often predict species presence at unsuitable depths, and AquaMaps maps often extrapolate species presence far afield from known occurrences. something about recommendations here? how this understanding can lead to better use of these datasets?

To understand the implications of these differences, we reexamined two recent global analyses that depend on these datasets: the Ocean Health Index and a global analysis of gaps in coverage of marine protected areas. how far to describe results here?


0.2 Significance

subset of the abstract


1 Introduction

Peer beneath the waves anywhere in the world - a moody kelp forest, a bustling coral reefscape, the frigid depths beneath a polar ice sheet, the endless blue of the open ocean. Which species would you expect to encounter, and which species would be woefully out of place? Mapping and predicting species distributions is fundamental to the sciences of ecology, biogeography, and conservation, among many others. Knowing where individuals of a species exist, and where they thrive, provides foundational information for understanding species ranges and diversity, predicting species responses to human impacts and climate change, and managing and protecting species effectively. A rich literature tackles the many dimensions of these questions.

One major outcome of this body of science is the various compiled databases of species distribution maps. Two important repositories predict marine species ranges throughout the world’s oceans – AquaMaps modeled species distribution maps (Kaschner et al. 2013) and International Union for Conservation of Nature (IUCN) species range maps (REF). These spatial datasets are used for a wide range of purposes, including assessing marine species status (Halpern et al. 2012, Selig et al. 2013), evaluating global biodiversity patterns (Coll et al. 2010, Martin et al. 2014), predicting range shifts (Molinos et al. 2015), and setting conservation priorities (Klein et al. 2015).

The two datasets ostensibly describe the same information, but significant differences in methodology and intent could lead to dramatically different understandings of our marine ecosystems, with significant implications for policy and conservation recommendations.

Importantly, biases in taxonomic or spatial coverage within a dataset could shift management or conservation actions away from places or species that are most in need.(maybe Jetz 2008?) Inaccurate indications of presence or absence could lead to inefficient or ineffective marine reserve systems and management plans. (Rondinini et al. 2006) (Jetz 2008?)

<<<<<<< HEAD

To understand the implications of differences between these two datasets, we compared how each data set represents the global spatial and taxonomic distribution of species. For the relatively small number of species mapped in both datasets, we examined how well the species maps align. We then subjected two recent marine biodiversity studies - the Ocean Health Index biodiversity goal (Halpern et al. 2012) and a global analysis of gaps in protection afforded by marine protected areas (MPAs) (Klein et al. 2015) - to a sensitivity analysis, substituting one dataset over the other, to highlight the possible consequences of different data use decisions on our understanding of the status of marine biodiversity.

=======

To understand the implications of differences between these two datasets, we compared how each data set represents the global spatial and taxonomic distribution of species. For the relatively small number of species mapped in both datasets, we examined how well the species maps align. We then subjected two recent marine biodiversity studies - the Ocean Health Index biodiversity goal (Halpern et al. 2012) and a global MPA gap analysis (Klein et al. 2015) - to a sensitivity analysis, substituting one dataset over the other, to highlight the possible consequences of different data use decisions on our understanding of the status of marine biodiversity.

>>>>>>> 2bde1ea0cf3f0e8312264b3c91b7b89420931187

2 Overview of AquaMaps and IUCN datasets

The IUCN publishes species range maps as spatial vector polygon shapefiles, bundled by taxonomic groups. Using GIS, experts outline spatial polygons to represent a given species’ extent of occurrence, based on observation records and refined by expert understanding of the species’ range and habitat preferences. IUCN releases a range map bundle for a taxonomic group once it has been “comprehensively assessed,” i.e. at least 90% of the species within the taxonomic group have been evaluated (REF). While this mitigates sampling bias within taxa, it also means that entire taxonomic groups remain unavailable until they have met this threshold of comprehensive assessment.

AquaMaps develops species distribution maps based on modeled relative environmental suitability. For each mapped species, environmental preferences (e.g. temperature, depth, salinity) are deduced from occurrence records, published species databases such as FishBase, and expert knowledge. The AquaMaps model overlays these environmental preferences atop a map of environmental attributes on a 0.5 degree grid, creating a global raster of probability of occurrence for each species.

IUCN range maps are intended to include all possible regions in which a species may be present, with the caveat that this does not imply the species is evenly distributed everywhere within the boundaries of the map(REF). AquaMaps modeled distribution maps provide a more nuanced prediction of species presence within the extent of occurrence, but the model may not capture all the complexities that drive species distribution (REF). Due to these differences in methodology and intent, IUCN range maps are more likely to overpredict species presence while AquaMaps modeled distribution maps are relatively more likely to underpredict species presence.(REF and REF)


3 Results and Discussion

3.0.1 Taxonomic distribution between datasets

<<<<<<< HEAD

Each dataset offers spatial distribution information for large numbers of species (22889 AquaMaps-mapped species; 4138 IUCN-mapped marine species). However, the datasets vary in terms of taxonomic coverage and regional coverage, with only 2166 species included in both datasets (9.5% of AquaMaps species; 52.3% of IUCN marine species; Figure 1a).

The distribution of IUCN-mapped species skews toward tropical latitudes and away from the Atlantic and Eastern Pacific compared to the distribution of AquaMaps-mapped species.(Figure 1b and 1c) This skew likely reflects the fact that the IUCN dataset focuses more heavily on coral reef-associated taxa than does the AquaMaps dataset (see Figure 1a).

=======

Figure 1(a): Number and proportion of species, listed by taxa, included in each dataset: IUCN, AquaMaps, or both.

Figure 1 (b, c): Global marine species richness according to (a) AquaMaps dataset and (b) IUCN dataset.

Each dataset offers spatial distribution information for large numbers of species (22889 AquaMaps-mapped species; 4138 IUCN-mapped marine species). However, the datasets vary in terms of taxonomic coverage and regional coverage, with only 2297 species included in both datasets (10% of AquaMaps species; 55.5% of IUCN marine species; Fig. 1, 2).

The distribution of IUCN-mapped species skews toward tropical latitudes and away from the Atlantic and Eastern Pacific compared to the distribution of AquaMaps-mapped species.(FIG) This skew likely reflects the fact that the IUCN dataset focuses more heavily on coral reef-associated taxa than does the AquaMaps dataset (see FIG).

>>>>>>> 2bde1ea0cf3f0e8312264b3c91b7b89420931187

For spatial assessments of biodiversity, the choice of one dataset over the other is likely to create significantly different results. For studies confined to a narrow range of taxa or to a narrow spatial scale, one dataset may offer an advantage over the other in the number of mapped species available. For global scale biodiversity studies, however, the selection of one dataset over the other will entail tradeoffs in spatial coverage, taxonomic breadth, and taxonomic depth.use something from Rondinini here?

Small overlap means that using both datasets in conjunction will add a huge number of species without duplicating efforts, if the differences between the two methods can be reconciled appropriately

find a reference that describes what makes a “good” dataset for global biodiversity, e.g. OHI, or maybe species richness vs diversity vs “health” or whatnot: (Tittensor et al., 2010)


3.0.2 Defining spatial alignment between the two datasets

The IUCN and AquaMaps spatial datasets share 2166 species in common. If we were to examine each of these species’ pair of maps side by side, we would hope to see spatial correlation both in the global pattern of species distribution (where on the map) and the extent of species range (how much of the map).

<<<<<<< HEAD

In 69.8% of cases, the IUCN range map indicated a larger species range than the AquaMaps map. This finding concurs with the general expectation that geographic range maps such as IUCN are more likely to over-predict presence than predicted distribution models such as AquaMaps, while predicted distribution models are more likely to over-predict absence.(Rondinini et al., 2006) errors of commission vs errors of omission, essentially type I and type II errors

Dividing the map-paired species into quadrants based on median values for each dimension, we can examine the implications of four different qualities of alignment. (Figure 2a)

Examining spatial alignment by taxonomic group (Figure 2b), we found that certain taxa were far more likely than others to be spatially well-aligned; in particular, wide-ranging pelagic organisms (50% or more of ocean area) such as marine mammals, tunas, and billfishes were more consistently well-aligned (quadrants 1 and 2) than demersal and reef organisms. propose an explanation here?

Quadrant 1 contains species whose map pairs agree in both spatial distribution and the extent of described ranges, as we would expect for well-understood species. Excellent and valuable, but not particularly interesting.

The extent-misaligned maps contained in quadrant 2 provide a richer set of examples to understand the fundamental differences between the datasets; examining these, we can identify several mechanisms that could cause such extent misalignment.

For 630 of 682 species in this quadrant (92.4%), the IUCN extent is larger than the AquaMaps extent. By itself, this is not surprising; but for many of these species, a quick look at the maps (see SOM for examples) shows that the IUCN range hews closely to the AquaMaps range, while adding a significant buffer zone. Coral species dominate this quadrant (294 coral species - 58.2% of all corals and 43.1% of all species in quadrant 2). We suspect that many of these extent-misaligned map pairs can be easily explained: most corals and reef-associated organisms prefer shallower waters; seafloor depth is explicitly modeled in AquaMaps, but not explicitly considered in IUCN range considerations. (For examples and details, see SOM)

  • verify some of the coral maps by saving raster, showing raster and polygons in QGIS against a bathymetry layer
  • include recommendation for IUCN range maps to be clipped to bathymetry where appropriate? or save til end?

The extent-alignment of species maps found in quadrant 3 can in many cases be attributed to “two wrongs make a right.” For these species, IUCN ranges are frequently overextended into unsuitable habitat, as in the case of many quadrant 2 species, but at the same time, AquaMaps aggressively extrapolates distribution of presence into locations where IUCN predicts absence. This results in an area ratio close to 100%, giving the impression of extent alignment. this quadrant seems to be a false alignment then?

Quadrant 4 species maps seem to suffer from the same problem as those in quadrant 3, but here the AquaMaps extrapolation extends even further, pushing the extent alignment ratio lower. In this quadrant, AquaMaps predicts a larger extent than IUCN for 70% of species represented in this quadrant, compared with just 30% of species in the entire set of overlapping maps (this is supposed to show that much of this quadrant is based on overextrapolation of AquaMaps).

  • Q4: more data-poor models and fewer reviewed species maps; data-rich models and expert review improve AquaMaps duh…
  • recommend filtering for occurcells >= 10, and where possible, giving extra weight to reviewed maps

table to show data poor status and reviewed status for AquaMaps maps represented in the quadrant plot

quadrant n species n data poor species n reviewed species
total 22889 8749 (38.2%) 1296 (5.7%)
paired 2166 457 (21.1%) 290 (13.4%)
q1 401 33 (8.2%) 100 (24.9%)
q2 682 151 (22.1%) 100 (14.7%)
q3 682 114 (16.5%) 65 (9.5%)
q4 410 159 (39.7%) 25 (6.2%)

data poor = fewer than 10 points used to generate the environmental envelopes; reviewed = map checked or formally expert-reviewed

=======

In 69.8% of cases, the IUCN range map indicated a larger species range than the AquaMaps map. This finding concurs with the general expectation that geographic range maps (e.g. IUCN) are more likely to over-predict presence than predicted distribution models (e.g. AquaMaps), while predicted distribution models are more likely to over-predict absence.(Rondinini et al., 2006) errors of commission vs errors of omission, essentially type I and type II errors

Figure 3 (a): For each paired-map species, we calculated two dimensions of spatial alignment: distribution alignment, which we defined as the proportion of the smaller range intersecting the larger range (where on the map); and extent alignment, which we defined as the ratio of the smaller range to the larger range (how much of the map). For a species whose distribution is well understood and described in both datasets, we would expect to see a value near 100% for each dimension of alignment.

The upper right quadrant (quadrant 1) comprises species whose maps largely agree (better than median value) in both spatial distribution and the extent of described ranges (n = 400; 18.5 %). The upper left quadrant (quadrant 2) comprises species whose maps agree well in distribution, but disagree in extent (n = 684; 31.6 %). The lower right quadrant (quadrant 3) includes species for which the paired maps generally agree in range extent, but disagree on where those ranges lie - a more worrisome mismatch than that indicated by quadrant 2 (n = 681; 31.4 %). The lower left quadrant (quadrant 4) indicates species for which the map pairs agree poorly in both area and distribution (n = 401; 18.5 %).

Dividing the map-paired species into quadrants based on median values for each dimension, we can examine the implications of four different qualities of alignment. (FIG 3a)

Examining spatial alignment by taxonomic group (FIG 3b), we found that certain taxa were far more likely than others to be spatially well-aligned; in particular, wide-ranging pelagic organisms (50% or more of ocean area) such as marine mammals, tunas, and billfishes were more consistently well-aligned (quadrants 1 and 2) than demersal and reef organisms. propose an explanation here?

Figure 3 (b): Spatial alignment of paired-map species by taxonomic group.

Quadrant 1 contains species whose map pairs agree in both spatial distribution and the extent of described ranges, as we would expect for well-understood species. Excellent and valuable, but not particularly interesting.

The extent-misaligned maps contained in quadrant 2 provide a richer set of examples to understand the fundamental differences between the datasets; examining these, we can identify several mechanisms that could cause such extent misalignment.

For 630 of 682 species in quadrant 2 (92.4%), the IUCN extent is larger than the AquaMaps extent. By itself, this is not surprising; but for many of these species, a quick look at the maps (see SOM for examples) shows that the IUCN range hews closely to the AquaMaps range, while adding a significant buffer zone. Coral species dominate this quadrant (294 coral species - 58.2% of all corals and 43.1% of all species in quadrant 2). We suspect that many of these extent-misaligned map pairs can be easily explained: most corals and reef-associated organisms prefer shallower waters; seafloor depth is explicitly modeled in AquaMaps, but not explicitly considered in IUCN range considerations. (For examples and details, see SOM) include recommendation for IUCN range maps to be clipped to bathymetry where appropriate? or save til end?

The extent-alignment of species maps found in quadrant 3 can in many cases be attributed to “two wrongs make a right.” For these species, IUCN ranges are frequently overextended into unsuitable habitat, as in the case of many quadrant 2 species, but at the same time, AquaMaps aggressively extrapolates distribution of presence into locations where IUCN predicts absence. This results in an area ratio close to 1, giving the impression of extent alignment. wrap up something with quadrants 3 and 4: In several instances, the AquaMaps map conformed to a small, local subset of the IUCN map (see SOM for examples - what was the Q4 graph?), in some cases due to some point locality observations being rejected by AquaMaps experts (EXAMPLES) and in other cases due to differences in species identification. bleh - this needs help. Evidence? examples?

Quadrant 4 species maps seem to suffer from the same problem as those in quadrant 3, but here the AquaMaps extrapolation extends even further, pushing the extent alignment ratio lower. In this quadrant, AquaMaps predicts a larger extent than IUCN for 70% of species represented in this quadrant, compared with just 30% of species in the entire set of overlapping maps.

how many species are “reviewed” in each quadrant? how many are data poor? overall: 21% of overlapping species are data poor (model based on fewer than 10 cell-occurrences) q1: 8.2% (33/401); q2: 22.1% (151/682); q3: 16.5% (114/692); q4: 39.7% (159/401) Overall: 13.4% of species reviewed… q1: 24.9%; q2: 14.7%; q3: 9.5%; q4: 6.2%

>>>>>>> 2bde1ea0cf3f0e8312264b3c91b7b89420931187

Other hypotheses/case studies to highlight other possible mechanistic differences?


4 Implications

4.1 Application to OHI

The global Ocean Health Index (OHI) (Halpern et al. 2012), an index made up of 10 goals, utilizes both of these datasets to inform the Species subgoal of the Biodiversity goal. As it is currently calculated, the Species subgoal uses species spatial distribution data and IUCN Red List conservation status to calculate an area-weighted mean species status in each of 221 exclusive economic zones. Spatial distributions were gleaned from both IUCN and AquaMaps datasets, preferring IUCN data for species represented in both data sets. OHI uses a probability threshold of 40% to determine presence for AquaMaps data. Species with no spatial data in either dataset were excluded, as were species with insufficient information to determine conservation status (including species listed as not evaluated or data deficient).

Briefly summarize results

Since the Ocean Health Index Species subgoal relies on spatial data from both datasets, the impacts of these threshold and preference changes will be somewhat muted. When IUCN range maps are the preferred data source, only the subset of AquaMaps-only species will be affected by threshold changes; and when AquaMaps is the preferred source, the IUCN-only species will dampen the effect of a threshold change. But


4.2 Application to MPA Gap Analysis

do the analysis…

Predictions: What happens when using a 0% threshold instead of a 50%? What happens when using IUCN rather than AM? IUCN overestimates extent esp for coastal species? More species will be included in MPAs so fewer apparent gap species. Included range area inside MPAs will increase. Policy implications?


4.3 Wrap it up

Recommendations on data use?

  • IUCN mapmakers
    • at least clip polygons to bathymetry for depth-sensitive species, come on now people
    • think carefully about quality of data in – e.g. are all OBIS points equally valuable? AquaMaps seems to disagree - but are they justified in doing so?
  • AquaMaps mapmakers
    • get expert eyeballs on all the maps
    • Explain rationale for discarding some of the data points - can there be a defined decision tool for this?
  • Users
    • understand that range maps tend to overestimate presence; AquaMaps are less likely to overestimate presence, but trade off for overestimating absence. Both better than simple point locality data.
    • to use both together:
      • consider setting AM presence threshold to 0% to create an extent of occurrence map, more like IUCN map.
      • consider additional filters and criteria, e.g. “reviewed” or “occurcells”
      • consider clipping IUCN ranges to bathymetry for depth-limited species or taxa
      • or consider de-weighting IUCN for area-weighted calcs, for taxa in which IUCN is consistently overestimating presence

5 Methods

5.0.1 taxonomic distribution comparison

To examine the overall taxonomic distribution across the spatial datasets, we grouped species by taxonomic class and data source (IUCN, AquaMaps, or both), and examined the proportion of each class represented in each data source category. We then filtered the species list to those that have been evaluated for the IUCN Red List of Threatened Species (in SOM).

Currently this is also included in body - take out and leave in here? Per BH: “Either cut or move to data description section of methods above.”: The IUCN releases spatial data sets when a taxonomic group (typically on the scale of order or family) has been comprehensively assessed, to guard against sample bias (though non-comprehensive datasets are available for reptiles and marine fish). As such, spatial data for many taxonomic classes remain unavailable, and within a class, the assessed sub-groups may not represent the entire class.

5.0.2 global spatial distribution comparison

To compare the spatial representation of the two datasets directly, we first rasterized the IUCN species polygons to the same 0.5° grid as the AquaMaps species maps; species presence within a grid cell was determined by any non-zero overlap of a species polygon with the cell, and species richness per cell was simply the count of the species present. For the AquaMaps dataset, we determined per-cell species richness by counting all species with non-zero probability of occurrence, to best approximate the “extent of occurrence” generally indicated by IUCN maps. We represented relative distribution of species richness for each dataset by plotting average species count against latitude and longitude.

map pairs comparison

Using genus and species binomials to identify paired maps, we selected the subset of marine species that have range maps in both IUCN and AquaMaps current native distribution (n = 2166). We used the same criteria as outlined above to determine species presence within a cell.

from results/discussion: Overlaying paired distribution maps for a given species, we calculated two dimensions of spatial alignment: distribution alignment, which we defined as the proportion of the smaller range intersecting the larger range; and area alignment, which we defined as the ratio of the smaller range to the larger range.

\[\alpha_{dist} = \frac{A_{small \cap large}}{A_{large}} * 100\%\]

\[\alpha_{area} = \frac{A_{small}}{A_{large}} * 100\%\]


6 Figures

<<<<<<< HEAD

6.0.1 Figure 1

Figure 1(a): Number and proportion of species, listed by taxa, included in each dataset: IUCN, AquaMaps, or both. AquaMaps encompasses a broader range of taxa than IUCN, while IUCN focuses on comprehensively assessing select taxonomic groups, typically at the level of order or family. Overlapping species are dominated by bony fishes (1304 species, primarily tropical taxa) and corals (505 species).

Figure 1 (b, c): Global marine species richness according to (a) AquaMaps dataset and (b) IUCN dataset. The frequency plot to the right of each map shows relative species count per cell at each latitude; while both datasets peak in tropical latitudes near the equator, the frequency for IUCN maps drops quickly beyond 30°N and 30°S, while the frequency for AquaMaps remains robust well into temperate latitudes. The frequency plot above each map shows relative species count at each longitude, showing a slight bias in the IUCN dataset away from the Atlantic and eastern Pacific compared to AquaMaps.

6.0.2 Figure 2

Figure 2 (a): For each paired-map species, we calculated two dimensions of spatial alignment: distribution alignment, which we defined as the proportion of the smaller range intersecting the larger range (where on the map); and extent alignment, which we defined as the ratio of the smaller range to the larger range (how much of the map). For a species whose distribution is well understood and described in both datasets, we would expect to see a value near 100% for each dimension of alignment.

The upper right quadrant (quadrant 1) comprises species whose maps largely agree (better than median value) in both spatial distribution and the extent of described ranges (n = 400; 18.5 %). The upper left quadrant (quadrant 2) comprises species whose maps agree well in distribution, but disagree in extent (n = 684; 31.6 %). The lower right quadrant (quadrant 3) includes species for which the paired maps generally agree in range extent, but disagree on where those ranges lie - a more worrisome mismatch than that indicated by quadrant 2 (n = 681; 31.4 %). The lower left quadrant (quadrant 4) indicates species for which the map pairs agree poorly in both area and distribution (n = 401; 18.5 %).

Figure 2 (b): Spatial alignment of paired-map species by taxonomic group.

6.0.3 Figure 3

=======

Keep figures in the text while working, for easier editing and captioning.
Put figures here once the manuscript is done

6.0.1 Figure 4

>>>>>>> 2bde1ea0cf3f0e8312264b3c91b7b89420931187

Absolute change in SPP score

Figure [X] shows the change in status score for the Species Subgoal within the global Ocean Health Index under three different scenarios.

Scenario Priority data source AquaMaps presence threshold
Scenario 0 (current) IUCN >= 40%
Scenario 1 IUCN > 0%
Scenario 2 AquaMaps >= 40%
Scenario 3 AquaMaps > 0%
  • Scenario 1 shows the effect of reducing the presence threshold for AquaMaps presence. Reducing the threshold will always increase the apparent range of a species, therefore the slight decrease in average score suggests increased spatial representation of threatened species.

  • Scenario 2 shows the effect of preferring AquaMaps data over IUCN, while maintaining the same presence threshold. This will have different effects depending on the species; in general, AquaMaps ranges are smaller than IUCN ranges, so many but not all overlapping species will see a decrease in represented range. The slight bump in mean score may indicate a small increase in spatial representation of low-risk species, a small decrease in spatial representation of high-risk species, or more likely a combination of both.

  • Scenario 3 shows the effect of preferring AquaMaps data over IUCN, while eliminating the presence threshold. Just as a presence threshold of zero in scenario 1 drives a decrease in average score relative to the baseline, the zero threshold in scenario 3 drives a decrease in scores relative to scenario 2. The large decrease seems to indicate that within the set of paired-map species, a zero threshold greatly increases the spatial representation of high-risk species relative to low-risk species.


7 References

Key AquaMaps publications:

Kaschner, K., R. Watson, A. W. Trites, D. Pauly (2006). Mapping world-wide distributions of marine mammal species using a relative environmental suitability (RES) model. Marine Ecology Progress Series 316: 285–310. check this citation journal name… This outlines the basic RES methodology - AM development

Kaschner, K., D.P. Tittensor, J. Ready, T Gerrodette and B. Worm (2011). Current and Future Patterns of Global Marine Mammal Biodiversity. PLoS ONE 6(5): e19653. PDF just what the title says - AM development, presence threshold 60%, also analyzes richness as a function of threshold

Ready, J., K. Kaschner, A.B. South, P.D Eastwood, T. Rees, J. Rius, E. Agbayani, S. Kullander and R. Froese (2010). Predicting the distributions of marine organisms at the global scale. Ecological Modelling 221(3): 467-478. PDF Presents AM; assessing AquaMaps against other presence-only species models

Papers based on AquaMaps:

Jones, M.C., S.R. Dyeb, J.K. Pinnegar and W.W.L. Cheung (2012). Modelling commercial fish distributions: Prediction and assessment using different approaches. Ecological Modelling 225(2012): 133-145. PDF comparison of species distribution models including AquaMaps, Maxent and the Sea Around Us Project

Coll, M., C. Piroddi, J. Steenbeek, K. Kaschner, F. Ben Rais Lasram et al. (2010). The biodiversity of the Mediterranean Sea: estimates, patterns, and threats. PLoS ONE 5(8): e11842. PDF used AquaMaps to predict Med biodiversity. Also: Threshold = 0.

Martin C.S., Fletcher R., Jones M.C., Kaschner K., Sullivan E., Tittensor D.P., Mcowen C., Geffert J.L., van Bochove J.W., Thomas H., Blyth S., Ravillious C., Tolley M., Stanwell-Smith D. (2014). Manual of marine and coastal datasets of biodiversity importance. May 2014 release. Cambridge (UK): UNEP World Conservation Monitoring Centre. 28 pp. (+ 4 annexes totalling 174 pp. and one e-supplement). PDF report on marine data sets and data gaps etc, incl both IUCN and AM


Hurlbert 2007 Species richness, hotspots, and the scale dependence of range maps in ecology and conservation. _mostly rasters of range maps? “The scale dependence of range-map accuracy poses clear limitations on braod-scale ecological analyses and conservation assessments. … we provide guidance about the approriate scale of their use_

Jetz 2008 Ecological Correlates and Conservation Implications of Overestimating Species Geographic Ranges EOO maps are usually highly interpolated and overestimate small-scale occurrence, which may bias research outcomes

Pimm 2014 The biodiversity of species and their rates of extinction, distribtuion, and protection. uses range maps to show biodiversity areas; may use IUCN range maps. Also discusses gaps and possible things that can be done about them.

Rondinini 2006 Tradeoffs of different types of species occurrence data for use in systematic conservation planning compares point locality, range maps, and distribution models in terms of omission and commission errors; also outlines Extent of Occurrence and Area of Occupancy distinctions.

García Molinos, Jorge, Benjamin S. Halpern, David S. Schoeman, Christopher J. Brown, Wolfgang Kiessling, Pippa J. Moore, John M. Pandolfi, Elvira S. Poloczanska, Anthony J. Richardson, and Michael T. Burrows. “Climate Velocity and the Future Global Redistribution of Marine Biodiversity.” Nature Climate Change advance online publication (August 31, 2015). doi:10.1038/nclimate2769.

Halpern, Benjamin S., Catherine Longo, Darren Hardy, Karen L. McLeod, Jameal F. Samhouri, Steven K. Katona, Kristin Kleisner, et al. “An Index to Assess the Health and Benefits of the Global Ocean.” Nature 488, no. 7413 (August 30, 2012): 615–20. doi:10.1038/nature11397.

Klein, Carissa J., Christopher J. Brown, Benjamin S. Halpern, Daniel B. Segan, Jennifer McGowan, Maria Beger, and James E.M. Watson. “Shortfalls in the Global Protected Area Network at Representing Marine Biodiversity.” Scientific Reports 5 (December 3, 2015): 17539. doi:10.1038/srep17539.

Selig, Elizabeth R., Catherine Longo, Benjamin S. Halpern, Benjamin D. Best, Darren Hardy, Cristiane T. Elfes, Courtney Scarborough, Kristin M. Kleisner, and Steven K. Katona. “Assessing Global Marine Biodiversity Status within a Coupled Socio-Ecological Perspective.” PLoS ONE 8, no. 4 (April 11, 2013): e60284. doi:10.1371/journal.pone.0060284.


8 Supplemental Information

8.1 Info on data prep

8.1.1 processing AquaMaps

  • start with .sql files - three of ’em - how to get ’em?
  • turn into .csvs, which columns are critical for this analysis? simplify

8.1.2 downloading and processing IUCN maps

  • which data sets are included?
  • raster::extract() to convert polys to csvs
  • which columns are included?

8.1.3 creating master species list for co-listed species

  • using AquaMaps and IUCN, and IUCN master list, create the big list
  • which columns are included?
  • adjust this master list to use IUCN SID only for parent - eliminate the whole question of parent/subpop? for the purpose of this analysis

8.2 descriptions of IUCN and AM data sets

IUCN: While the polygons roughly define regions of presence/absence, additional attributes provide information on extant/extinct ranges, native/introduced ranges, and seasonality.

As of December 2015, IUCN had published species distribution maps for 4138 marine species across 24 taxonomic groups. For this analysis, we did not consider IUCN range maps for bird species, as those data are hosted separately by BirdLife International.

As of December 2015, AquaMaps current native distribution maps have been produced for 22889 species.

Move to supplementary materials. - BH: For example, as of this writing, IUCN has released no spatial data for class Elasmobranchii (cartilaginous fishes including sharks and rays); and while IUCN offers a large number of maps within class Actinopterygii (ray-finned bony fishes), the available maps include only a few primarily tropical taxonomic sub-groups, such as wrasses, damselfish, butterflyfish, tunas, and billfishes, but are missing economically important subgroups including salmon, rockfish, and clupeids. However, IUCN’s criterion of comprehensive assessment greatly reduces the risk of sample bias within the bounds of the assessed taxonomic groups.

Probably just cut, but maybe include in Suppl Materials. - BH: The release of AquaMaps distribution maps is not limited to comprehensively-assessed taxa, and maps are available across a much larger range of taxonomic classes; however, there is no guarantee that the list of species included within each class is a representative cross section of the entire class.

Red List inclusion:

  • All IUCN-mapped species are also included in Red List species, but only 30% of AquaMaps species.

  • Breaking down the quadrants by IUCN extinction risk categories (FIG 3c), we found that species with higher extinction risk tend to be better aligned between the two datasets, perhaps correlated to increased expert scrutiny. Does higher perceived risk lead to increased attention, and thus better understanding of species distribution? Or conversely, does increased attention to species distribution reveal more species at risk? Likely both mechanisms are at play on a case-by-case basis, depending on the species’ taxon and region. does this argument bear up to closer scrutiny? CR isn’t dominated by Q1 any more

8.3 illustrative maps for different quadrants and different mechanistic problems

a few codes to indicate likely problems?

  • DC = depth clipping; IUCN extents go beyond reasonable depth for the species
  • DX = data excluded; AM excludes some observations that appear to be used for IUCN
  • DP = data poor (AquaMaps acknowledges data poor status? or use less than 10 data points as criteria)
  • NX = needs evaluation by expert; AM model likely predicts areas where species is unlikely to be found based on observed occurrences - does AM model reflect same as AM suitable habitat?

Q2:

species quad group codes notes
Conus episcopatus q2 cones DP, DX d-match around Coral Triangle, but IUCN shows much more around Indian Ocean and S Pacific; also IUCN extents
Ctenochaetus binotatus q2 IUCN extents beyond AM
Naso vlamingii q2 acanthuridae IUCN extents beyond AM
Thalassoma purpureum q2 wrasses IUCN extents beyond AM
Porites nigrescens q2 corals 2 DP AM limited to CT; IUCN shows far greater dist, and extents
Conus ammiralis q2 cones DP AM
Conus tessulatus q2 cones extents
Acropora sarmentosa q2 corals 1 extents; also AM shows HI but IUCN does not?
Holothuria fuscogilva q2 sea cucumbers extents; also AM shows Mex and Central Am west coast, but IUCN does not (not available online?)
Oculina Varicosa q2 corals 2 lim to Caribbean; extents - this might be a good close-up map
Acanthocybium solandri q2 tunas/billfish mostly tropical by IUCN; AM shows much farther N and S

Q3

species quad group codes notes
Conus magnificus q3 cones IUCN distribution is more limited than AM distribution)
Abudefduf concolor q3 damselfish Baja for AM, southern Central Am for IUCN; maybe a good closeup map? AM points from Baja to Peru incl Galapagos; some rejected pts from Caribbean and eastern Atlantic
Centropyge aurantonotus q3 angelfish AM shows Caribbean; IUCN shows northern South America, also maybe a good closeup map? AM points from FL to south Brazil, but no Gulf of Mexico.
Chlorurus perspicillatus q3 parrotfish IUCN shows only HI; AM shows dots all over S Pacific and near Marianas Islands
Sarda orientalis q3 tunas/billfish limited dists by IUCN
Montastraea franksi q3 corals 2 DP IUCN shows just in Caribbean; AM shows all up & down east coast of N America and S America

Q4

species quad group codes notes
Acanthurus nigroris q4 IUCN dist limited to HI, while AM shows all over S Pacific and CT - AM observations all over S pacific…
Praealticus tanegasimae q4 blenny IUCN shows limited dist around Japan/Marianas; AM shows much broader through CT and W Pacific - good local map?
Stethojulis marquesensis q4 wrasse IUCN shows only small cluster of islands (Marquesas?) while AM shows much broader across S Pacific
Canthigaster leoparda q4 puffer IUCN limited spots around a couple of islands; AM shows all throughout CT
pentacheles snyderi q4 lobster DP AM shows all over - sim to depth profiles? while IUCN shows only a couple patches - HI? S Pacific, and S Indian oceans? AM limits to specific FAO regions, which results in odd cutoffs; also shows some data points in N Atlantic which has an odd result.
Nephropsis sulcata q4 lobster IUCN shows odd ranges, most of which don’t overlap with AM

8.4 AquaMaps: Effect of changing “presence” threshold on apparent distribution

8.4.1 AquaMaps presence threshold analysis - move to SOM

For our comparisons of global distribution of represented biodiversity and spatial alignment between datasets, we considered “present” to be any cell with a non-zero probability of occurrence, to best approximate the “extent of occurrence” as generally indicated by IUCN maps. To examine the effect of different presence threshold selections on the represented range of a species, we varied the threshold from 0.05 to 1.00 and calculated the average species range relative to a zero threshold.

8.4.2 Figure 5 (a, b):

AquaMaps distribution map extent remaining after applying a presence threshold. (a) A 40% threshold applied to all species in the AquaMaps dataset shows a mean loss of XXX, with a wide distribution in which some species lose nearly all of their apparent range. (b) Mean (median) remaining extent at increments of presence threshold. Dark grey ribbon includes 25% to 75% quantiles, while light grey ribbon includes 9% to 91% quantiles.

AquaMaps distribution maps indicate “probability of occurrence” within each 0.5° cell, with values ranging from zero to one, rather than a simple present/absent value as indicated by IUCN maps. Many studies convert this AquaMaps probability to a simple presence value by assigning a threshold value (REF references here). A higher threshold constrains an analysis to cells with near certainty of occurrence, while a low threshold captures larger areas of increasingly marginal suitability.

At a presence threshold of 40%, as used in the Ocean Health Index Species subgoal, the bulk of AquaMaps species suffer a significant decrease in represented range, and some species lose nearly their entire range. Incrementing the presence threshold from 0.00 to 1.00 for the entire AquaMaps dataset, the shallow downward trend indicates a low but consistent sensitivity to threshold choice, with no surprising tradeoffs that could suggest an “optimal” threshold. This pattern may not hold true for all subsets of AquaMaps species, however, whether subsetting by taxa or by georegion.